speaker-0: Hi everyone and welcome to the BobotixTack podcast. Today we first get into LionTix Spatial and we will learn about what they have been up to since releasing Pokemon Go. After that, we get into an update on the robot hand market. A lot has been happening over the past few weeks. Then we get into robot boyfriend and girlfriend. It's a topic that has been around for a while, but lately some companies have been releasing new robots. And finally we get into the United IPO, obviously. My name is Leo.
speaker-1: My name is Oscar and let's get started. Yes, Niantic Spatial. Super interesting company because they are trying to build essentially the Google maps for robots, but in 3D and with eyes. They are working on something which is called a large geospatial model. So an LGM. So not a world model. This is a persistent machine readable model of the real world. And we will get into it, but first of course, let's dive into the history. They started over 20 years ago. two thousand one with a company called Keyhole. And this one was founded by John Hank and the CTO Brian McClanton. And this was pretty much the technology that later on fed into Google Earth because it was acquired in 2004. And then the founders subsequently worked at Google, especially in mapping. And then later on, Niantic Labs, this was not Niantic Spatial, was incubated inside of Google in 2011. And then spun out in twenty fifteen. And that was pretty much the Niantic that we know, the one that was building games such as Pokemon Go, for example. I don't know if you've played it, Leo. Have you tried the game when it came out?
speaker-0: Pokémon Go? Yeah, of course. Like it was like 10 years ago or something, right?
speaker-1: Yeah, it's long ago, but still it was at that time a pretty advanced game if you remember. Yeah. And they had to solve quite a few problems. For example, understanding the real world locations, because it was an AR game, you know? Aligning digital content with the physical landmarks, camera localization, under changing viewpoints. And obviously all these things can be relevant to robotics as well. So it already kinda started there. In twenty twenty five, they then had some sort of a corporate break because Scopely, a gaming company, agreed to buy Niantic's gaming business for $3.5 billion. So this included Pokemon Go, Pikmin Gloom, and a couple of other games that they had. And Niantic pretty much sold them and on the same day Niantic Spatial Inc. was launched as an independent company with $250 million in funding. So they pretty much said okay let's get rid of the gaming business. We believe in this geospatial AI. And we will make a sole company for this purpose. SNAP, for example, invested in them as part of a multi-year strategic partnership. As of right now, they had a little CEO change in March. So before it was John Hank, the founder from Keho, and Brian McLedon is also still there as the CTO. But in March, In Chosu became the CEO. And then we have also Thomas Jeweki, who is the president and COO. So they sort of made a really strong restart here. Now, what are the products that they are working on? Well, the main strategic centerpiece that they have is the large geospatial model, so LGM. And this is pretty much an attempt to create a shared persistent geospatially anchored model of the physical reality. And of course, they have a huge data foundation for that. A bit from the games and also a couple of other sources, and we will get to know more of their sources as well. But they have around 30 billion post real-world images or even more than that, and over 300 million user-contributed scans. So that's a huge amount of data that they have. And for the images, for example, post is important because these are not just random images, these are actually associated with the camera position and the orientation. So quite a deep amount of data that they have there. And why does this matter for robotics? Well, robotics right now, or most robots as of today. They maintain a combination of different systems basically that they have. They have an onboard visual odometry, so the robot estimates movement using its cameras, motion, and the sensors, of course, or wheels. Then they can use SLAM, which is simultaneous localization and mapping. So here the robot essentially builds a map and keeps track of it. Or we can also have systems like GPS in place. But often these type of systems are robot or fleet specific. and have to be locally maintained. So their vision is basically bringing all this together and have a shared spatial memory. So if a robot enters an environment, it inherits the prior geometric and semantic knowledge. So it doesn't start from zero all the time. Now as I said before, it's not a world model. A world model is dreaming, generating plausible environments. And this is different. It's grounded in the geospecific truth. So to give you an example for an input, You could show it the image of a street and then it could tell you, okay, you are at this location and you're facing in this direction. Or you give it a scene and you ask it, okay, where can the robot drive safely? And it could output you a traversable region or a route that the robot can take. And they also specify that this is not to replace world models or VIAs or anything else. They more see it as an addition to that that can be used. And also we have to be clear that this is not finished. It's not a finished model. They're still building this. But there are already parts of the stack and the data pipeline. For example, they have Niantic VPS 2.0, the 2.0 version just launched in April. And VPS stands for visual positioning system. So this is pretty much what I described. It determines a camera's position and orientation by matching what the camera sees against the previously mapped information. And also it works without pre-scanning since the 2.0 version. So GPS, for example, tells you you are approximately here. And VPS can tell you, okay, from these bricks, from this window, from this pavement, and from all these visual features that it perceives, it can tell you you are here and you're facing this direction. It's kind of like GeoGuesser if you know the game. Do you know GeoGesser? So it's a little bit like that. And they are
speaker-0: Yeah, of course.
speaker-1: Really trying to optimize their data pipeline. For example, they have a product called Scaniverse, and this is pretty much a consumer-friendly 3D scanning product. So you can take your phone and you can start scanning stuff. You can also use 360 cameras and then you can basically do 3D scanning with it. And it's pretty advanced because it can actually export a Gaussian splat together with an aligned collision mesh. In USDZ. And this is obviously useful if you want to later on use this in NVIDIA ISIC sim for simulation, robot simulation, and things like that. They even say that just using a five minute 360-degree capture can give you a simulation ready environment. This means they drastically reduce the cost for real to sim with these technologies that they are providing. And a lot of people are using them already.
speaker-0: Just allow me to specify one thing maybe. I made a few posts on Twitter about that recently, so it's still pretty fresh in my in my mind. ⁓ maybe for people who don't know, ⁓ what we can specify is that a Goshen splat is like a fast low compute way of making a 3D model of the world. The only issue with that is you don't get the collision mesh that Niantic also offers. The collision mesh is much more compute intensive traditionally to make. But I'm guessing in this case they found a way to make it low compute since it goes with the Gaussian spladding, right?
speaker-1: Yes, one hundred percent correct. That's that's a really good addition. They really try to make it easy for anyone to use Scaniverse, you know, but also companies as well. And they have a lot of different partnerships. And one is the one with Flexion and NVIDIA. And if you remember we talked about Flexion, you talked about it in one of our podcasts before. And they just recently did a collaboration. So NVIDIA was also in the game and they demonstrated a really useful workflow for real to sim and then sim to real for humanoid robot learning. So this was in July this year. Niantic Spatial essentially produced a high fidelity digital twin of an office with their software and their tools that they have and their technology. Nvidia supplied the simulator and the GPU accelerated training infrastructure with ISEC sim, you know, ISIC lab, the tools that they have for simulation learning. And then Flexion was there to pretty much develop and adapt the robot policies. and the software deployment. And together they trained an RGB only navigation policy entirely inside a reconstructed office, that the one that was reconstructed from Liantic, and transferred it zero shot to a real robot. And that's obviously super nice. I mean they reported that it successfully navigated the office and outperformed a dev-based policy. Self-reported benchmark again, not 100% production ready probably, but it just demonstrates a really useful And also potentially scalable workflow. Because imagine you go to BMW, you have a BMW plant, and you know you just scan the entire plant and then you simulate in that plant. And obviously you want to have the results of that simulation as close to the results that you will later on achieve when you actually deploy in reality. And they're kind of getting closer with this approach. So really interesting partnership there. They have another one with Coco Robotics and you talked about surf robotics as well, you know, these little delivery robots essentially. And Coco Robotics is a pretty similar company. And Niantic is collaborating with them, giving them their VPS or their positioning system and and stuff like that. Then they are also collaborating with Vantor. I don't know if you know Vantor, but they are in the satellite data business. And together with them they are developing a common air-to-ground coordinate system so that drones and robots and vehicles can remain spatially aligned. For example, when GPS is not available. They're also doing something with the Human Machine Collaboration Institute and the city of Rancho Cordova in California, where they put built a digital twin of an entire city. So this is this is pretty new, but it's crazy. They're doing a digital twin of the entire freaking city and create a machine readable city. Well that's their goal at least. And ultimately this is super useful because you can let robot companies, for example, train and simulate against this. And of course eventually deploy in the corresponding physical city with all this previous work that you have done. And hopefully this results in higher success rates and better deployment workflows. They also partner with Spexi for drone imagery. They connect Spexi's drone capture network with Niantics reconstruction API and then reconstruction APIs you use some data, for example, that you took with Scannyverse and then you reconstruct it. They still have their partnership with Snap, obviously. They want to expand on an AI powered real world map. So I think the picture is pretty clear, right? They want to build this model and they have all these partnerships and companies and real life touch points where they can get data, you know, and then ultimately bring it into this model. They have their own app, their own software where you can scan, but they also have these partners which provide them different types of data, satellite data, whatever they need, basically. I think it's a really interesting approach that they have. The only thing that came to my mind was obviously, okay, where is Google? Because you know they ultimately came from Google. So and Google owns Google Maps and Google Earth, you know, these are really advanced products as well. And anybody watching this podcast has seen these cars driving around and scanning. So Google the stakeholder there as well, but this was just an open question for me. And obviously the model is not ready. And physics, we don't really know how much they incorporate into this model.
speaker-0: I see. Very insightful. I'm very curious about ⁓ what's gonna happen with them.
speaker-1: They seem to have really good people on board, definitely.
speaker-0: ⁓ yeah, yeah, definitely. From from the beginning. They have really strong ingenios for sure. And I'm guessing it's the right time to get into our second topic of the day, which is the hand market. And an overall update on it. So for the viewers who are just discovering us, we have previously talked about hands in the podcast in a previous episode, where we covered the main hand types and the existing hands at the time. Direct drive hands and also tendon-driven hands. Which are the two main types, there are also a few others. We are going to cover a new hand type that has just been released and talk about startups that are in the space. Because basically, hands are a considerable part of a humanoid robot when it comes to the amount of engineering you need to put in it. Right now there is like the preliminary phase of the robot games in China. There is like a hundred meter dash trial, you know, robots competing on a hundred meter run. But there is no juggling trial. It shows you that running is easy for robots right now, but ⁓ handling things with a hand is not as easy at all. The reason why is because human hands have twenty-seven degrees of freedom and if you want to build a robot hand it has to have at least nineteen to twenty three degrees of freedom in order to be credible from an anthropomorphic point of view. For example, generation two, Tesla Optimus, the humanoid robot, his hands were about nine thousand five hundred dollars. More than 17% of the total cost of the robot. Because you have to make all the other components you use in the robot, you have to make them smaller with better heat dissipation and overall better engineering to fit into such a small space. And another factor is that, for example, in the automotive industry, the typical mean time to failure, which is how how long does a part work before you have to replace it or replace a part of it, is around 50,000 hours. And right now. Mean time to failure data for humanoid robots and therefore robot hands is non-existent. And if you look at Stanford AI index in 2026, robots have succeeded at roughly 12% of household tasks. Household tasks are like the perfect benchmark for dexterity manipulation. It's mostly about taking things with your hand, manipulating objects with you know more or less force depending on the task. So one of the first company I want to mention today is called Tacta System. They are from Palo Alto in California and they came out of Telse on July 27. And maybe you have seen them around. Yeah. They have raised 75 million dollars in total, 11 million seed while they were still in stealth, and 64 million Series A, which they announced when they got out of stealth pretty much. Executive chairman is Andreas Bible, who sold companies to Apple and Fujifilm, and CEO is Vikram Pavati, and he sold a company. to Lockheed Martin. So both are strong veterans of the industry. And what's really interesting about their product is that they ship two different hands right now. One of them is a three-finger hand and the other is a typical five finger hands. They also mentioned that if you want, they can make a six or seven finger hand for you. You know, there is no problem because the design is modular. So if you need more finger, you can have it. Each of their fingers has three independently actuated joints. So the number of degrees of freedom Obviously depends on the number of fingers you have in that hand. And what they did at first is that they put tactile data categorphs on expert factory workers, mostly in Asia, so Japan, China, Vietnam, Taiwan, and a little bit in Mexico as well. And they recorded that most of the high value precision work was only using three fingers, which is specifically these three fingers, so thumb index and middle fingers, and that the last two are kind of you know. Bagging along and most of the time you will use these two with your middle finger to add more strength. But if you are like, I don't know, screwing something or doing some precision part, pick and play, stuff like that, you will mostly use the first three fingers of your hand. So they say, okay, why do we need a five finger hand? Then maybe we don't. And so they started designing that three finger hands. What I like about it is that it's an entirely data driven decision. ⁓ it's not just a random vision or a random take. And the steel sheet five finger hands because the industry is still very much centers on five fingers. But there is another reason why everybody is making ⁓ you know five fingers is that most robot models are trained first for pre-training. They are trained on egocentric video data. Imagine you have a camera on your forehead and you record your point of view, seeing your hands, manipulate objects and make tasks. Your hands in this case have five fingers, so the data automatically Has five fingers where they do hand pose estimation. That's the main reason why the industry has converged on five fingers instead of three. Tacti is making the decision of shipping three also because it is cheaper, it is less moving parts, less maintenance. And also, if you remember, a few weeks ago we also talked about Sunday. And Sunday also has a three finger hand. Sunday also offers a data capture glove that matches exactly the hand of the robot. And the way Sunday does it as a reminder is that they have Sum, they have the index, and then the middle ring and pinky are together as one strong large finger, right? So it's another way to make three fingers and I'm guessing if you have a five-finger hand that works for you and you want to use tact as model, you could just say, okay, the you know the ring finger and pinky finger are just going to tag along with the middle finger, stick with it and Then your five finger hands just become a three finger hand with one very large finger at the end. I guess that could work. What's also interesting about Tacta's hand is that they are building it around an entirely new paradigm. What Tacta does is fluidic tendons, which is a new architecture. ⁓ they are not exactly the first one to try that, but they are the first one to commercialize it at scales, for sure. So inside their hands there are cables with fluid inside them that drives pistons that then move the finger. I'm gonna quote some of the CEO's claim because obviously we don't really have independent feedback on it. But Pavati claimed that it allows to make the finger very, very thin, obviously, ⁓ thinner than if they were direct drive or you know traditional tendon. There is no heat generated anywhere close to the hand because you can put the motors as far back as in the shoulder or in the torso. So no heat in the most moving part of the hand is a considerable advantage global.
speaker-1: 'Cause Yeah, that's one advantage of the orbit actuator from Bryce as well, right?
speaker-0: Yeah, yeah, definitely. Heat dissipation is a real engineering problem in robotics, for sure. And he also said that it is more precise and also that it is more reliable because if you need to service or maintain the hand, you can do it just by accessing the motors which are outside of the hand. You don't necessarily need to unpack the entire hand to maintain it. And I also found out that a company called Sanctuary AI has worked for years on fluidic attrition as well. But they recently pivoted to a software only product and they kind of seems to have either ditched their product or put it on the back burner. So I'm not really sure as Tacta solved problems that ⁓ Sanctuary has not. And if Tacta manages to make it, then it's going to be a really strong advantage for them. Tacta has glove for capture, you can buy the glove. They are rated for 30 million cycles. They have figure tip sensors for touch sensitivity and force. And they are replaceable and explicitly sold at AC Consumable because obviously they are in contact with, you know, all the surfaces and objects you are going to manipulate. More importantly, they have false data in their capture gloves that they can sense and recreate with a tactile sensor at the fingertip of the hand. They are aiming for the first customer's deployment in early twenty twenty seven. And I really like this company. I think their hands look really cool. They seem to be extremely strong and precise. So I'm really good to see what's next for them. Another company that we have to mention is called Xinova, and it's a Chinese company. And like everything in China, it does something similar, just much faster. So it was founded in Hangzhou at the end of twenty twenty four and they kept rising ever since. They raised an engine round in December twenty twenty five, Series A in May twenty twenty six, and a Series A plus in July twenty twenty six. In total they raised close to two hundred and thirty million dollars, which is considerable especially for a Chinese company. An American engineer is paid two point five times what a Chinese engineer is paid. And China has a really strong culture of corporate household, nine nine things, things like that. So people on average work a bit more. So let's just say that 230 million raised in China would be at least twice that raised in the US when it comes to how much you can spend on hiring employees. Their hand is called the Flex 2. It has 23 degrees of freedom, which is interesting because their Flex One previous product had twenty-five degrees of freedom and they kind of realize that you actually don't need, you know, too much. They have hybrid actuation, so a mix of cable driven And direct drive. Not exactly sure where does each stop and begin. Hybrid accretion seems to be like taking the best of both worlds, that it's becoming a bit more mainstream overall in ⁓ the hand market. It can do two fist extensions per second, which is, I don't know, interesting. It's also the first time I see such a metric being specified by a hand manufacturer. Also, they have a perfect score of the Carpanji Sum opposition test, which is Kind of a test the doctor makes you do like this, use your thumb to reach the tip of your other fingers, anything like that. It is pretty much a test that determines if you have a disability that hinders you from using your hand correctly. And again, it is the first time I see a robot hand manufacturer specifying it. And they have multimodal sensing, they have embedded cameras in the hand. The first version it was in the palm. And now in the second version it is in the wrist. It is obviously more comfortable to have the camera there. Because if you grab something it could hide or damage the camera, it can grasp 12 kilos per hand, four kilograms in continuous grasping. They have recently built a new facility targeting the release of 10,000 hands per year. And now if we take a little step back and consider China, China has been pretty early at having robot hands companies entirely. In the US or in the West, we also have, but they came in later and we have much less of them. China has Linkabot valued at three billion. Baksini core valued at one point five billion. Brainco valued at one point three billion and AG Link valued at one billion. So just China has, you know, four unicorns and I'm guessing Xenoa is probably going to become one as well very soon if they keep going like this. Among this we should spend a little bit of time on Edgilink because they shipped 8000 Vexter sands. And 10,000 grippers. And they have 1,000 units deployed in warehouse pharmacies and factories. 5kilo payload, and the founder comes from 10 cent robotics. Because quite often in the US you see, ⁓ yeah, this guy comes from Google, this guy comes from Tesla or you know another famous company. And you kind of to see this startup mafia windfall existing in China as well, which I find very interesting. A few more companies that I was mentioning: Inspire Robots. So this sheet the RH twenty six, which has only six active degrees of freedom. They say it covers roughly sixty to seventy of the human grasping taxonomy. It is the most widely integrated commercial hands in Chinese humanoids apparently. And it is also cheaper than ⁓ most of the other hands you can find on the market. So Tesla Optimus V three has twenty-two degrees of freedom and about twenty-five accurators that were moved into the forearms, not into the hand. So it is currently the state of the art or state of the industry, if we could put it this way, for tendon-based hands, which you move the accreditors in the forearms so the hand does not get as hot as it would for direct drive. But as the Tacta CEO says, ⁓ it's even better if you can move the accurator further away, like Tacta does. ⁓ figure O three is an interesting case because They do not believe, especially Brad Alcock, he doesn't believe in tendon-based hands. He believes in direct drives. He specifically said that doing research and development in tendon hands was a mistake for figure in the past. And so they are going all into direct motor actuations. The difficulty with that beyond heat is that you have to make extremely small and efficient actuators. And this is no small engineering feat, it is an actual hard problem. Unfortunately, they don't release specs beyond you know marketing lingo, so we don't really know how it works and we don't have that many demos either. ⁓ but it's still interesting to see that one of the main humanoid manufacturers in the US is betting on direct drive while nobody else is doing it anymore. I think it's something we have to follow closely. And another company I found which is interesting is called Gzilla Robotics and they make tactile skin for sensing and for sensitivity that you can put on any robot hands you have. So it's like a retrofit to provide sensitivity to existing hands.
speaker-1: No, it's interesting that you mentioned tactile skin because this is probably more sensor related tactile skin you're talking about, not like biomimetic in terms of that it looks and feels like the skin of a human. Yeah, exactly. Because obviously if we go into the direction of robot boyfriends and girlfriends, there are completely different problems that these companies face, of course. I mean they have kind of the same problems that a traditional robotics company has, but they also want to have a face. Because they pretty much wanna create social companions, okay? The extreme is of course having a relationship with a robot. I mean that's crazy, but given the development and given also, you know, a lot of people on the planet who are single and don't have a partner, at some point it might become an option for them, right? I mean, there are people who have married like a piece of Eiffel Tower or something like that. A robot wouldn't surprise me.
speaker-0: Yeah, and more and more people are having like deep intimate conversation with ChatGPT. Let's say if you can have Chat GPT embedded into a human with robots, I would not surpr I would not be surprised if it were to happen. Like
speaker-1: said. Yeah, because it's only getting better. And there are actually a couple companies really trying to build a companion. One of them is UbiTech, and they just released the U World U1 series. And UbiTech we have not really talked about them before, but they have a huge lineup. It's one of the major humanoid companies in China. Their newest model is basically a major move towards consumer companionship. And They're pretty much positioning it like a lifelike humanoid for emotional support and potentially cyber boyfriend girlfriend use cases. They have a couple different models. So one is the U1 Lite, so this is semi-torso. Then they have the U1 Pro, which is full body, and the U1 Ultra, which is the most advanced one when it comes to full body movement. And the price starts at 17,000 while the ultra configurations go up to 130. And 146,000. So it's pretty interesting. They have a lot of different technological features, up to 88 degrees of freedom, silicon, biomimetic skin, expressive facial and eye movement. And this is super interesting if we look at, you know, Tesla's robot, Figures robot, or also some other humanoid robots, they don't even have a face, right? They just have the
speaker-0: I would say they have a glass mask that hides the camera behind it.
speaker-1: Yeah, that's accurate. And yeah, and this one also comes with as you mentioned an emotionalware LM recognizing over twenty emotional states. So it can tell when you cry, when you laugh, and all these kind of things. It's supposed to come with long term memory as well, and roughly five hundred milliseconds of interaction response time and largely local processing for this latency of course. UBTEC reported because this has just been released already Over thirteen thousand orders at launch, but it's not independently verified. So we will see when they get delivered and how this plays out. But you can see videos of these robots. They look kind of realistic. I mean you can still tell it's not a human, of course, but they are really pushing the human like realism there. And you can actually buy them right now. I didn't find them on their non-Chinese site. You cannot buy them there, but they're like third party vendors where you can buy them. So that's UBTech. Another company which has been in this space for a while is a Real Botics and they produce highly realistic AI humanoids as well. One example is their model called ARIA and it combines silicone, face, facial tracking, also conversational AI, and roughly forty motorized degrees of articulation. So the face takes a lot of engineering work for them obviously because that's where the magic happens, right? If you talk to someone and they don't respond to you at all then There's no connection. But you can already buy it. There's an entry level that starts at around $20,000, full body one at around 125,000. So these are actually more expensive than even state-of-the-art humanoids sometimes. And what's interesting about them, because they kind of started with this romantic narrative, I think, but then the company structure changed. And robotics explicitly only focuses on commercial and service humanoids. So you know, more like professional. Human-like appearance, not really this romance or companionship type of thing. And that's why they separated the business into Intima. So Intima LIC became a separate company from them for adult wellness and companionship. And Intima, on the other hand, owns 86% stake in the real doll business. And Real Doll is also some sort of long-running company that Ships these AI enhanced silicon companions. They also have like customizable bodies. They have a proprietary skin technology, for example, but they're still pretty heavy and not really that mobile. I wouldn't call it a walking humanoid yet. Also, it's a very niche market because these are quite expensive. They said in June 26 they were actively shipping a batch of 12 units. Then we have another company which is called Ahead Form, and that is a Chinese startup. And they're developing One of the most visually human-like robotic hats and upper bodies. So they're solely focused on that, but they really do great work. They have some amazing videos on their website and they have a couple different models already. One is called the Origin F1, then the Origin M1, and then the ELF HUAN. And there's also a 2.0 version of that. And the Origin F1 has 26 facial degrees of freedom, multimodal vision and voice, face recognition. So imagine you really talk to a robot. With another person maybe and then it says like, Hey look, it's you Leo. Hey look, that's you, Oscar. You know, I can talk to you, it can differentiate, and then maybe even has memory for each person and it can relate to that. It can be a really deep connection there. They have something called the Origin M1 as well, which has 30 movable facial muscle units, micro expressions, eye tracking, and skin advertised around human body temperature. So we they really want to maintain skin that it has actually the temperature of a human. So it feels more natural. Then they have a elf, which is called Shuan, with also over 30 facial degrees of freedom. And they tease a 2.0 version on the website already. It's more like a inquiry reservation type of thing rather than just like a transparent pricing. So it's probably around 50 to 150,000 for these. And then we have Zomnia Lab. And these are quite new. And I don't really know how legit they are, but it's kind of like a new model that they are working on, the Model L. And L stands for love machine. And this is presented as a full-size intimacy robot whose personality, memory, and relationship are intended to develop over time. So this is really going for this full fantasy. Okay, I'm pretty much marrying a robot right now. They have some provisional specification, which is approximately 170 centimeters, 22 kilograms, so it's quite a light, a carbon fiber skeleton, 22 degrees of freedom. Soft compliant joints, pressure, contact sensing, localized warmth, customizable appearance. Now, what's interesting about them is they have a website and there's like a blurred layer on top of the actual website content. So I'm not sure how much of this content is just what I was able to scrape using an LM and how much is actually legit. So I just want to say that. But apparently they have a couple different models and they will take deposits for reservations from around $300 for the Model L. 500 for the Pro and 2000 for the Pro Max. So that's another one. Then we have Love Sense AI doll, and this is a companion doll. Much more affordable at around $4,000 to $8,000. Full silicon body, approximately $155 to $159. 33 kilograms, articulated posability, facial movement, touch sensing, persistent conversational memory. You know, all these features that we already talked about. They mostly go for the same things, obviously. But this one is fundamentally more a posable AI intimacy doll than like a real humanoid. We have another couple mentions. One is Jigglyjoy X Robots. It's a company working on this lifelike skin. So they really try to get skin to feel exactly like human skin. We have a couple of different Chinese doll makers that add AI. And then interestingly, I also found out about a cyber brothel ⁓ in Berlin. It's actually here in Berlin.
speaker-0: How convenience.
speaker-1: I went on their website and I looked at it and it's basically yeah, a brothel but they have these dolls, you know, they have these kind of dolls and people are apparently going there and having a lot of fun with them. So it's already taking place. I mean whoever is like still debating this, it will come and maybe there will probably be more people going for this type of situation ship or relationship. Who who knows, you know?
speaker-0: Yeah, definitely. Let let me know if you visit the Berlin establishment and you can make a vlog out of it.
speaker-1: Yeah, I'll I'll take you with me and then we we will have a podcast there with ⁓ with the founder maybe.
speaker-0: That could be actually an interesting thing. Yeah. I I'm curious about when will we have the first IPO of, you know, a robot companion startup. The day it happens, ⁓ we will definitely have massive adoption of robots in our daily lives, even beyond the companionship thing, I think.
speaker-1: I mean UbiTech is pretty big. I'm not sure if they are had already an IPO.
speaker-0: ⁓ yeah, they had a they had an IPO, UBTEC.
speaker-1: ⁓ yeah. So that's one. I mean then we got one already.
speaker-0: Yeah, actually UbiTech is already public, but their share kind of went down a little bit after Unitries IPO. I think UbiTech was not in the robot companion business at first, right? They were a traditional robotics company.
speaker-1: Yeah. I mean they just released this whole entire new section.
speaker-0: It's a new thing for them. So maybe Unitree will get into it as well. I don't know. We will see. It doesn't seem to be part of their plan, but anything is possible in this space and everything moves super fast. And it's actually a good way to talk about the Unitree IPO that you have probably heard about. It's been literally everywhere. So Unitree priced its share at twenty two point four dollars on August sixth. And they went public on August nineteenth, which is yesterday at the time of this recording, and they were aiming to raise nine hundred and four million dollars. And what happened is that when the markets opened, the company was up six hundred and thirty percent. So are massively oversubscribed. It was valuing the company at sixty-six billion dollars. Then it went down a little bit and it closed its first day up four hundred and eighty eight percent. at a fifty three billion dollars valuation. So yeah, I mean it's a ballpark, right? Because depending on how you look and the exchange rate, some sources say like it's fifty one billion, some other fifty three billion dollars, but you you get the idea. It's ⁓ it's an over fifty billion dollars valuation in a public market. If we look at multipliers, it's two hundred and ten times their sales, one thousand three hundred times their earnings, which is massive.
speaker-1: Not bad. So
speaker-0: You know, it's open AI anthropic kind of level of private VC valuation because none of this company is public yet. They were oversubscribed ⁓ more than eight thousand times, which means if you were connecting to your train desk and wanted to buy a unitary share, you were getting one in five thousand and five hundred and twenty-six chance of winning one. The stock was so oversubscribed that they went from offering 20% to retail investors to offering 30% in the end. So yeah, you had to get pretty lucky just to get one. ⁓ who profited the most from the IPO? Obviously it's Wong Xing Zing, the CEO and founder, and also Mae Chuan, which is ⁓ very large Chinese tech companies that does delivery. And they made more than 70x what they invested in Unitry, more than five billion dollars. Overall. If you have to compare the Star Market Composite, which is the Chengai market aggregate, was off six point wide percent. At the same time, Unitry was up five hundred percent. We have to talk a little bit about their products. What do they do exactly? So they shipped a bit more than five thousand humanoids in twenty twenty five. It is the company in the world that shipped the most. And we are expecting much stronger numbers in 2026, not only from Unitry, but also from other companies. They went from selling less than 10 in 2023. To around 400 in 2024 to 5,000 in 2025. So that's a considerable growth. Revenue share went from 1% to 51%. And in dollars it's approximately $130 million per year. They also sell a lot of quadruplets. They have been doing quadruplets for nine years actually. And they sold more than thirty-three thousand. It is more than forty percent of their main business, approximately fifty-seven percent of their growth margin. So Humanoids is what gets the hype. Quadrupeds are what makes their business stable and real world. And they also make components actually. Even though in 2022 components were 18% of their business, it fell down to 6% in 2025. So I'm guessing Unitree doesn't want to sell components that much anymore. They just keep doing it because they make everything in-house. Because like I said, they make every critical subcomponent themselves and they have their in-house quasi direct drive, which is an actuator made. of a motor paired with a low ratio planetary gearbox. And most of the industry otherwise is more in favour of the strain wave gearbox. Unitree builds also harmonic reducers, but most of their joints are you know crazy direct drive. The reason for that is they are faster to make and faster to iterate over from an engineering perspective. And they are back driveable, which means if you control let's say a robot arm with such an iTrader, you can yourself push it the other way and you know the motor will be fine, which is not possible with other type of actuals. The caveats of this is that they make a lot of heat and they consume a high amount of poles. Obviously the two go hand in hand, right? Heat distribution is, you know, an issue like we mentioned with Tacta. And also in the previous episode we talked about someone performing surgery with two unit free robots. And one of the issues they had, even though the attempt was successful, is that robots were overheating quite a lot. So if you look at the history of the evaluation, on November 2, 2023, the Ministry of Industry and Information Technology in China issued the guiding opinions on the innovation and development of humanoid robots. So it's basically a piece of public Chinese industrial policy. And they were directed at Developing the market and manufacturing for humanoid robots in the country. The targets were in 2025 they had to have a preliminary innovation system, breakthroughs in the brain, cerebellum, and limbs, whatever that means, secure supply of core components, batch production at scales, and two, three globally influential ecosystem enterprise and two to three industrial clusters. And I'm guessing they already have that because you mentioned Ubitech, which is Chinese. already public manufacturing humanoids unitree and when it comes to industrial cluster I don't know where UbiTech is from but Unitree is from Hengzhou in China close to Shanghai and ⁓ there is also a strong cluster in humanoid robots in Shenzhen obviously so their targets were successfully reached in this case.
speaker-1: Yeah, they ⁓ Ub Tak is in change it.
speaker-0: Mm, makes sense. So yeah, we have Hangzhou and Shenzhen for clusters. By twenty twenty-seven it was also aiming at having two modules integrated into manufacturing supply chain. And ⁓ this is already something we see in China. We've seen the edgy bot stream in a factory, and this was earlier this year in twenty twenty-six. This is definitely happening by twenty twenty-seven. And what's interesting is that before that, in twenty fifteen, they had already made robotics one of the ten priority sectors in 2016 to 2020, they made a robot industry development plan and in 2021 the five year robotics plan put further emphasis into developing the robotic industry in China. And Unitary really rode that wave perfectly when it comes to getting public funding. Because the Ministry of Industry and Information Technology in China set a ten thousand unit human deployment target by end of twenty twenty six. And since Unitary alone is shipping more than five thousand, They are probably close to target, but we don't know for sure. And in China there is a really strong central government, but also giving a lot of freedom to all the states and local region, which all came up with their own public fund. And that's what Unitary took advantage of. Unitree had a pretty modest seed and they were largely unknown outside of quadruple robot research and locomotion. Basically everything robot leg related, right? Then They had Series A after the MIIT guidance, Series B in twenty twenty four, Series C in late twenty-four and early twenty twenty-five in two parts because they had a part with public money and a part with private fund. Apparently you cannot mix the both. And now they have an IPO at fifty-three billion. So it means that in 14 months their valuation went 30x, while their revenue went three hundred and thirty-five percent up. So You can see the pattern where there is like a central state designation of the priority. The provincial funds then invest in the relevant companies. And after that, private capital follows the signal. And we definitely see that with Unitary right now, live. We've seen that with BYD, for example, in the past, but it is becoming even more clear with Unitary. For sure. What's interesting is that Agility Robotics is looking apparently at a spec at around four billions with
speaker-1: Yeah, those are crazy numbers.
speaker-0: Around 300 million committed in pre-orders and multi-site deployments. They haven't released numbers on profitability or revenue, but we can see that Unitree closed its IPO at 13 times agility's valuation, right? Another interesting thing is who actually buys one? For humanoids, it's mostly research and education at 74%. After that, the rest is around commercial and consumer industry application. And ⁓ even though Humanoid revenue is clearly tied to defined industry or job because later down the line that's where the revenue will come from. See my analysis estimate that there are 250 Unitary humanoids in active labor application against more than 5,000 shit in 2025. And Unitary has apparently a 32.4% global market share in the humanoid market. And they still have a market share of over 60% on quadruped. Pretty strong position. Well, it's already oversubscribed, so I don't think I can right now, but maybe later. I don't know. But Unitree is a profitable company with over 30 million dollars in profit last year. So this is also why I think everybody is like betting on it. They announced what they intend to do with the money they raised through the IPO. So three hundred million dollars will go to embodied large models, roughly a hundred million per year.
speaker-1: So you're gonna buy some unit stock?
speaker-0: So Reunitary is definitely investing heavily into world models and robot embedded models. They don't want just to make the hardware, they also want to make the software. And they have an interesting architecture ⁓ running you know a VLA and a world plus action model in parallel, but they haven't said too much on it at this time. It's you know news from earlier announcements. They are also investing a hundred and sixty four million in ⁓ robotics body research and development. Also investing a bit less in new products and manufacturing base. And since the IPO was oversubscribed, they made they raise more money than they were intended. So they have 254 million unallocated right now. So let's see what they do with that. Maybe robot companion department. I don't know. And also recently, right before the IPO, Unitree released the Superman demo that you have probably seen. Which is like a demo of a United robot jumping super high. like a base vertical jump and also running at approximately twelve point six meters per second, which is more than Houston Bolt, for example. Again, nothing is independently verified. It's a company marketing video right before the IPO. The robot has no hands, no payload, but the robot games are happening, so we'll get actual numbers on a hundred meters dash later on. And it is still a pretty impressive you know feat of engineering. And it shows that unitary is not just Exiting with the IPO and then being done with it, they are just doubling down on R and D and getting into what's next for humanoids. ⁓ also we mentioned that Unitree was part with NVIDIA of the Isaac growth reference humanoids. But basically Nvidia ships the software and Unitree ships the hardware to get like a baseline humanoid to do research on. It ships with a Unitree H2 and a lot of Nvidia hardware for embedded compute. and NVIDIA software, but since then there has been the import ban and they have not communicated on what's happening with this project and how can it you know go further. I'm guessing Nvidia is going to partner with US companies and Unitree investing into developing its own models. So I'm guessing the two are going to kind of split up. A little bit more about the world humanoids robot game. So they are running between August twenty two and August twenty sixth at Beijing National Speed Skating Oval Stadium. They have more than six hundred teams, more than two thousand robots and sixty six countries across six continents, and fifty-one events, and eight of the events are data rest hands competition, which I find pretty interesting. What's also really cool this year is that the hundred meter dash is fully autonomous, no teleoperation. So again I'm really curious to see what kind of time are the robots able to make on this. And more importantly, are they going to go faster than the fastest humans ever, who is use and bolt, right? It makes me wonder a little bit overall is that does Unitree need the US market? Because for example, ⁓ DGI, which is the brand of our microphone right now, is banned from the US and they keep, you know, ⁓ going around this band by rebranding in weird and funny ways. And at the same time, the US is thirteen percent of Unitree's business. So do they really need it or would for them just focusing on Europe, Asia, Africa? Would be enough for them. ⁓ I don't know what's your take on that.
speaker-1: I mean, I've I've absolutely no idea, but I saw someone in the comments writing that there's he learned from some shop who's already starting like manufacturing robots in the US. And I'm not really sure if there's gonna be like some licensing deal going on and to take more control. And I've absolutely not verified this claim so far, but I think this looks like an interesting approach that I could see working, some sort of licensing and then local manufacturing.
speaker-0: Yeah, yeah, definitely. So yeah, it was a fun episode to research and robotics just keeps getting exciting as a space. So thank you everyone for watching. Please consider subscribing to our channel and to our Twitter. And we will see you again next week in the Robotics Tag podcast. My name is Leo. Bye bye. See ya.
speaker-1: My name is Oscar.
speaker-1: My name is Oscar and let's get started. Yes, Niantic Spatial. Super interesting company because they are trying to build essentially the Google maps for robots, but in 3D and with eyes. They are working on something which is called a large geospatial model. So an LGM. So not a world model. This is a persistent machine readable model of the real world. And we will get into it, but first of course, let's dive into the history. They started over 20 years ago. two thousand one with a company called Keyhole. And this one was founded by John Hank and the CTO Brian McClanton. And this was pretty much the technology that later on fed into Google Earth because it was acquired in 2004. And then the founders subsequently worked at Google, especially in mapping. And then later on, Niantic Labs, this was not Niantic Spatial, was incubated inside of Google in 2011. And then spun out in twenty fifteen. And that was pretty much the Niantic that we know, the one that was building games such as Pokemon Go, for example. I don't know if you've played it, Leo. Have you tried the game when it came out?
speaker-0: Pokémon Go? Yeah, of course. Like it was like 10 years ago or something, right?
speaker-1: Yeah, it's long ago, but still it was at that time a pretty advanced game if you remember. Yeah. And they had to solve quite a few problems. For example, understanding the real world locations, because it was an AR game, you know? Aligning digital content with the physical landmarks, camera localization, under changing viewpoints. And obviously all these things can be relevant to robotics as well. So it already kinda started there. In twenty twenty five, they then had some sort of a corporate break because Scopely, a gaming company, agreed to buy Niantic's gaming business for $3.5 billion. So this included Pokemon Go, Pikmin Gloom, and a couple of other games that they had. And Niantic pretty much sold them and on the same day Niantic Spatial Inc. was launched as an independent company with $250 million in funding. So they pretty much said okay let's get rid of the gaming business. We believe in this geospatial AI. And we will make a sole company for this purpose. SNAP, for example, invested in them as part of a multi-year strategic partnership. As of right now, they had a little CEO change in March. So before it was John Hank, the founder from Keho, and Brian McLedon is also still there as the CTO. But in March, In Chosu became the CEO. And then we have also Thomas Jeweki, who is the president and COO. So they sort of made a really strong restart here. Now, what are the products that they are working on? Well, the main strategic centerpiece that they have is the large geospatial model, so LGM. And this is pretty much an attempt to create a shared persistent geospatially anchored model of the physical reality. And of course, they have a huge data foundation for that. A bit from the games and also a couple of other sources, and we will get to know more of their sources as well. But they have around 30 billion post real-world images or even more than that, and over 300 million user-contributed scans. So that's a huge amount of data that they have. And for the images, for example, post is important because these are not just random images, these are actually associated with the camera position and the orientation. So quite a deep amount of data that they have there. And why does this matter for robotics? Well, robotics right now, or most robots as of today. They maintain a combination of different systems basically that they have. They have an onboard visual odometry, so the robot estimates movement using its cameras, motion, and the sensors, of course, or wheels. Then they can use SLAM, which is simultaneous localization and mapping. So here the robot essentially builds a map and keeps track of it. Or we can also have systems like GPS in place. But often these type of systems are robot or fleet specific. and have to be locally maintained. So their vision is basically bringing all this together and have a shared spatial memory. So if a robot enters an environment, it inherits the prior geometric and semantic knowledge. So it doesn't start from zero all the time. Now as I said before, it's not a world model. A world model is dreaming, generating plausible environments. And this is different. It's grounded in the geospecific truth. So to give you an example for an input, You could show it the image of a street and then it could tell you, okay, you are at this location and you're facing in this direction. Or you give it a scene and you ask it, okay, where can the robot drive safely? And it could output you a traversable region or a route that the robot can take. And they also specify that this is not to replace world models or VIAs or anything else. They more see it as an addition to that that can be used. And also we have to be clear that this is not finished. It's not a finished model. They're still building this. But there are already parts of the stack and the data pipeline. For example, they have Niantic VPS 2.0, the 2.0 version just launched in April. And VPS stands for visual positioning system. So this is pretty much what I described. It determines a camera's position and orientation by matching what the camera sees against the previously mapped information. And also it works without pre-scanning since the 2.0 version. So GPS, for example, tells you you are approximately here. And VPS can tell you, okay, from these bricks, from this window, from this pavement, and from all these visual features that it perceives, it can tell you you are here and you're facing this direction. It's kind of like GeoGuesser if you know the game. Do you know GeoGesser? So it's a little bit like that. And they are
speaker-0: Yeah, of course.
speaker-1: Really trying to optimize their data pipeline. For example, they have a product called Scaniverse, and this is pretty much a consumer-friendly 3D scanning product. So you can take your phone and you can start scanning stuff. You can also use 360 cameras and then you can basically do 3D scanning with it. And it's pretty advanced because it can actually export a Gaussian splat together with an aligned collision mesh. In USDZ. And this is obviously useful if you want to later on use this in NVIDIA ISIC sim for simulation, robot simulation, and things like that. They even say that just using a five minute 360-degree capture can give you a simulation ready environment. This means they drastically reduce the cost for real to sim with these technologies that they are providing. And a lot of people are using them already.
speaker-0: Just allow me to specify one thing maybe. I made a few posts on Twitter about that recently, so it's still pretty fresh in my in my mind. ⁓ maybe for people who don't know, ⁓ what we can specify is that a Goshen splat is like a fast low compute way of making a 3D model of the world. The only issue with that is you don't get the collision mesh that Niantic also offers. The collision mesh is much more compute intensive traditionally to make. But I'm guessing in this case they found a way to make it low compute since it goes with the Gaussian spladding, right?
speaker-1: Yes, one hundred percent correct. That's that's a really good addition. They really try to make it easy for anyone to use Scaniverse, you know, but also companies as well. And they have a lot of different partnerships. And one is the one with Flexion and NVIDIA. And if you remember we talked about Flexion, you talked about it in one of our podcasts before. And they just recently did a collaboration. So NVIDIA was also in the game and they demonstrated a really useful workflow for real to sim and then sim to real for humanoid robot learning. So this was in July this year. Niantic Spatial essentially produced a high fidelity digital twin of an office with their software and their tools that they have and their technology. Nvidia supplied the simulator and the GPU accelerated training infrastructure with ISEC sim, you know, ISIC lab, the tools that they have for simulation learning. And then Flexion was there to pretty much develop and adapt the robot policies. and the software deployment. And together they trained an RGB only navigation policy entirely inside a reconstructed office, that the one that was reconstructed from Liantic, and transferred it zero shot to a real robot. And that's obviously super nice. I mean they reported that it successfully navigated the office and outperformed a dev-based policy. Self-reported benchmark again, not 100% production ready probably, but it just demonstrates a really useful And also potentially scalable workflow. Because imagine you go to BMW, you have a BMW plant, and you know you just scan the entire plant and then you simulate in that plant. And obviously you want to have the results of that simulation as close to the results that you will later on achieve when you actually deploy in reality. And they're kind of getting closer with this approach. So really interesting partnership there. They have another one with Coco Robotics and you talked about surf robotics as well, you know, these little delivery robots essentially. And Coco Robotics is a pretty similar company. And Niantic is collaborating with them, giving them their VPS or their positioning system and and stuff like that. Then they are also collaborating with Vantor. I don't know if you know Vantor, but they are in the satellite data business. And together with them they are developing a common air-to-ground coordinate system so that drones and robots and vehicles can remain spatially aligned. For example, when GPS is not available. They're also doing something with the Human Machine Collaboration Institute and the city of Rancho Cordova in California, where they put built a digital twin of an entire city. So this is this is pretty new, but it's crazy. They're doing a digital twin of the entire freaking city and create a machine readable city. Well that's their goal at least. And ultimately this is super useful because you can let robot companies, for example, train and simulate against this. And of course eventually deploy in the corresponding physical city with all this previous work that you have done. And hopefully this results in higher success rates and better deployment workflows. They also partner with Spexi for drone imagery. They connect Spexi's drone capture network with Niantics reconstruction API and then reconstruction APIs you use some data, for example, that you took with Scannyverse and then you reconstruct it. They still have their partnership with Snap, obviously. They want to expand on an AI powered real world map. So I think the picture is pretty clear, right? They want to build this model and they have all these partnerships and companies and real life touch points where they can get data, you know, and then ultimately bring it into this model. They have their own app, their own software where you can scan, but they also have these partners which provide them different types of data, satellite data, whatever they need, basically. I think it's a really interesting approach that they have. The only thing that came to my mind was obviously, okay, where is Google? Because you know they ultimately came from Google. So and Google owns Google Maps and Google Earth, you know, these are really advanced products as well. And anybody watching this podcast has seen these cars driving around and scanning. So Google the stakeholder there as well, but this was just an open question for me. And obviously the model is not ready. And physics, we don't really know how much they incorporate into this model.
speaker-0: I see. Very insightful. I'm very curious about ⁓ what's gonna happen with them.
speaker-1: They seem to have really good people on board, definitely.
speaker-0: ⁓ yeah, yeah, definitely. From from the beginning. They have really strong ingenios for sure. And I'm guessing it's the right time to get into our second topic of the day, which is the hand market. And an overall update on it. So for the viewers who are just discovering us, we have previously talked about hands in the podcast in a previous episode, where we covered the main hand types and the existing hands at the time. Direct drive hands and also tendon-driven hands. Which are the two main types, there are also a few others. We are going to cover a new hand type that has just been released and talk about startups that are in the space. Because basically, hands are a considerable part of a humanoid robot when it comes to the amount of engineering you need to put in it. Right now there is like the preliminary phase of the robot games in China. There is like a hundred meter dash trial, you know, robots competing on a hundred meter run. But there is no juggling trial. It shows you that running is easy for robots right now, but ⁓ handling things with a hand is not as easy at all. The reason why is because human hands have twenty-seven degrees of freedom and if you want to build a robot hand it has to have at least nineteen to twenty three degrees of freedom in order to be credible from an anthropomorphic point of view. For example, generation two, Tesla Optimus, the humanoid robot, his hands were about nine thousand five hundred dollars. More than 17% of the total cost of the robot. Because you have to make all the other components you use in the robot, you have to make them smaller with better heat dissipation and overall better engineering to fit into such a small space. And another factor is that, for example, in the automotive industry, the typical mean time to failure, which is how how long does a part work before you have to replace it or replace a part of it, is around 50,000 hours. And right now. Mean time to failure data for humanoid robots and therefore robot hands is non-existent. And if you look at Stanford AI index in 2026, robots have succeeded at roughly 12% of household tasks. Household tasks are like the perfect benchmark for dexterity manipulation. It's mostly about taking things with your hand, manipulating objects with you know more or less force depending on the task. So one of the first company I want to mention today is called Tacta System. They are from Palo Alto in California and they came out of Telse on July 27. And maybe you have seen them around. Yeah. They have raised 75 million dollars in total, 11 million seed while they were still in stealth, and 64 million Series A, which they announced when they got out of stealth pretty much. Executive chairman is Andreas Bible, who sold companies to Apple and Fujifilm, and CEO is Vikram Pavati, and he sold a company. to Lockheed Martin. So both are strong veterans of the industry. And what's really interesting about their product is that they ship two different hands right now. One of them is a three-finger hand and the other is a typical five finger hands. They also mentioned that if you want, they can make a six or seven finger hand for you. You know, there is no problem because the design is modular. So if you need more finger, you can have it. Each of their fingers has three independently actuated joints. So the number of degrees of freedom Obviously depends on the number of fingers you have in that hand. And what they did at first is that they put tactile data categorphs on expert factory workers, mostly in Asia, so Japan, China, Vietnam, Taiwan, and a little bit in Mexico as well. And they recorded that most of the high value precision work was only using three fingers, which is specifically these three fingers, so thumb index and middle fingers, and that the last two are kind of you know. Bagging along and most of the time you will use these two with your middle finger to add more strength. But if you are like, I don't know, screwing something or doing some precision part, pick and play, stuff like that, you will mostly use the first three fingers of your hand. So they say, okay, why do we need a five finger hand? Then maybe we don't. And so they started designing that three finger hands. What I like about it is that it's an entirely data driven decision. ⁓ it's not just a random vision or a random take. And the steel sheet five finger hands because the industry is still very much centers on five fingers. But there is another reason why everybody is making ⁓ you know five fingers is that most robot models are trained first for pre-training. They are trained on egocentric video data. Imagine you have a camera on your forehead and you record your point of view, seeing your hands, manipulate objects and make tasks. Your hands in this case have five fingers, so the data automatically Has five fingers where they do hand pose estimation. That's the main reason why the industry has converged on five fingers instead of three. Tacti is making the decision of shipping three also because it is cheaper, it is less moving parts, less maintenance. And also, if you remember, a few weeks ago we also talked about Sunday. And Sunday also has a three finger hand. Sunday also offers a data capture glove that matches exactly the hand of the robot. And the way Sunday does it as a reminder is that they have Sum, they have the index, and then the middle ring and pinky are together as one strong large finger, right? So it's another way to make three fingers and I'm guessing if you have a five-finger hand that works for you and you want to use tact as model, you could just say, okay, the you know the ring finger and pinky finger are just going to tag along with the middle finger, stick with it and Then your five finger hands just become a three finger hand with one very large finger at the end. I guess that could work. What's also interesting about Tacta's hand is that they are building it around an entirely new paradigm. What Tacta does is fluidic tendons, which is a new architecture. ⁓ they are not exactly the first one to try that, but they are the first one to commercialize it at scales, for sure. So inside their hands there are cables with fluid inside them that drives pistons that then move the finger. I'm gonna quote some of the CEO's claim because obviously we don't really have independent feedback on it. But Pavati claimed that it allows to make the finger very, very thin, obviously, ⁓ thinner than if they were direct drive or you know traditional tendon. There is no heat generated anywhere close to the hand because you can put the motors as far back as in the shoulder or in the torso. So no heat in the most moving part of the hand is a considerable advantage global.
speaker-1: 'Cause Yeah, that's one advantage of the orbit actuator from Bryce as well, right?
speaker-0: Yeah, yeah, definitely. Heat dissipation is a real engineering problem in robotics, for sure. And he also said that it is more precise and also that it is more reliable because if you need to service or maintain the hand, you can do it just by accessing the motors which are outside of the hand. You don't necessarily need to unpack the entire hand to maintain it. And I also found out that a company called Sanctuary AI has worked for years on fluidic attrition as well. But they recently pivoted to a software only product and they kind of seems to have either ditched their product or put it on the back burner. So I'm not really sure as Tacta solved problems that ⁓ Sanctuary has not. And if Tacta manages to make it, then it's going to be a really strong advantage for them. Tacta has glove for capture, you can buy the glove. They are rated for 30 million cycles. They have figure tip sensors for touch sensitivity and force. And they are replaceable and explicitly sold at AC Consumable because obviously they are in contact with, you know, all the surfaces and objects you are going to manipulate. More importantly, they have false data in their capture gloves that they can sense and recreate with a tactile sensor at the fingertip of the hand. They are aiming for the first customer's deployment in early twenty twenty seven. And I really like this company. I think their hands look really cool. They seem to be extremely strong and precise. So I'm really good to see what's next for them. Another company that we have to mention is called Xinova, and it's a Chinese company. And like everything in China, it does something similar, just much faster. So it was founded in Hangzhou at the end of twenty twenty four and they kept rising ever since. They raised an engine round in December twenty twenty five, Series A in May twenty twenty six, and a Series A plus in July twenty twenty six. In total they raised close to two hundred and thirty million dollars, which is considerable especially for a Chinese company. An American engineer is paid two point five times what a Chinese engineer is paid. And China has a really strong culture of corporate household, nine nine things, things like that. So people on average work a bit more. So let's just say that 230 million raised in China would be at least twice that raised in the US when it comes to how much you can spend on hiring employees. Their hand is called the Flex 2. It has 23 degrees of freedom, which is interesting because their Flex One previous product had twenty-five degrees of freedom and they kind of realize that you actually don't need, you know, too much. They have hybrid actuation, so a mix of cable driven And direct drive. Not exactly sure where does each stop and begin. Hybrid accretion seems to be like taking the best of both worlds, that it's becoming a bit more mainstream overall in ⁓ the hand market. It can do two fist extensions per second, which is, I don't know, interesting. It's also the first time I see such a metric being specified by a hand manufacturer. Also, they have a perfect score of the Carpanji Sum opposition test, which is Kind of a test the doctor makes you do like this, use your thumb to reach the tip of your other fingers, anything like that. It is pretty much a test that determines if you have a disability that hinders you from using your hand correctly. And again, it is the first time I see a robot hand manufacturer specifying it. And they have multimodal sensing, they have embedded cameras in the hand. The first version it was in the palm. And now in the second version it is in the wrist. It is obviously more comfortable to have the camera there. Because if you grab something it could hide or damage the camera, it can grasp 12 kilos per hand, four kilograms in continuous grasping. They have recently built a new facility targeting the release of 10,000 hands per year. And now if we take a little step back and consider China, China has been pretty early at having robot hands companies entirely. In the US or in the West, we also have, but they came in later and we have much less of them. China has Linkabot valued at three billion. Baksini core valued at one point five billion. Brainco valued at one point three billion and AG Link valued at one billion. So just China has, you know, four unicorns and I'm guessing Xenoa is probably going to become one as well very soon if they keep going like this. Among this we should spend a little bit of time on Edgilink because they shipped 8000 Vexter sands. And 10,000 grippers. And they have 1,000 units deployed in warehouse pharmacies and factories. 5kilo payload, and the founder comes from 10 cent robotics. Because quite often in the US you see, ⁓ yeah, this guy comes from Google, this guy comes from Tesla or you know another famous company. And you kind of to see this startup mafia windfall existing in China as well, which I find very interesting. A few more companies that I was mentioning: Inspire Robots. So this sheet the RH twenty six, which has only six active degrees of freedom. They say it covers roughly sixty to seventy of the human grasping taxonomy. It is the most widely integrated commercial hands in Chinese humanoids apparently. And it is also cheaper than ⁓ most of the other hands you can find on the market. So Tesla Optimus V three has twenty-two degrees of freedom and about twenty-five accurators that were moved into the forearms, not into the hand. So it is currently the state of the art or state of the industry, if we could put it this way, for tendon-based hands, which you move the accreditors in the forearms so the hand does not get as hot as it would for direct drive. But as the Tacta CEO says, ⁓ it's even better if you can move the accurator further away, like Tacta does. ⁓ figure O three is an interesting case because They do not believe, especially Brad Alcock, he doesn't believe in tendon-based hands. He believes in direct drives. He specifically said that doing research and development in tendon hands was a mistake for figure in the past. And so they are going all into direct motor actuations. The difficulty with that beyond heat is that you have to make extremely small and efficient actuators. And this is no small engineering feat, it is an actual hard problem. Unfortunately, they don't release specs beyond you know marketing lingo, so we don't really know how it works and we don't have that many demos either. ⁓ but it's still interesting to see that one of the main humanoid manufacturers in the US is betting on direct drive while nobody else is doing it anymore. I think it's something we have to follow closely. And another company I found which is interesting is called Gzilla Robotics and they make tactile skin for sensing and for sensitivity that you can put on any robot hands you have. So it's like a retrofit to provide sensitivity to existing hands.
speaker-1: No, it's interesting that you mentioned tactile skin because this is probably more sensor related tactile skin you're talking about, not like biomimetic in terms of that it looks and feels like the skin of a human. Yeah, exactly. Because obviously if we go into the direction of robot boyfriends and girlfriends, there are completely different problems that these companies face, of course. I mean they have kind of the same problems that a traditional robotics company has, but they also want to have a face. Because they pretty much wanna create social companions, okay? The extreme is of course having a relationship with a robot. I mean that's crazy, but given the development and given also, you know, a lot of people on the planet who are single and don't have a partner, at some point it might become an option for them, right? I mean, there are people who have married like a piece of Eiffel Tower or something like that. A robot wouldn't surprise me.
speaker-0: Yeah, and more and more people are having like deep intimate conversation with ChatGPT. Let's say if you can have Chat GPT embedded into a human with robots, I would not surpr I would not be surprised if it were to happen. Like
speaker-1: said. Yeah, because it's only getting better. And there are actually a couple companies really trying to build a companion. One of them is UbiTech, and they just released the U World U1 series. And UbiTech we have not really talked about them before, but they have a huge lineup. It's one of the major humanoid companies in China. Their newest model is basically a major move towards consumer companionship. And They're pretty much positioning it like a lifelike humanoid for emotional support and potentially cyber boyfriend girlfriend use cases. They have a couple different models. So one is the U1 Lite, so this is semi-torso. Then they have the U1 Pro, which is full body, and the U1 Ultra, which is the most advanced one when it comes to full body movement. And the price starts at 17,000 while the ultra configurations go up to 130. And 146,000. So it's pretty interesting. They have a lot of different technological features, up to 88 degrees of freedom, silicon, biomimetic skin, expressive facial and eye movement. And this is super interesting if we look at, you know, Tesla's robot, Figures robot, or also some other humanoid robots, they don't even have a face, right? They just have the
speaker-0: I would say they have a glass mask that hides the camera behind it.
speaker-1: Yeah, that's accurate. And yeah, and this one also comes with as you mentioned an emotionalware LM recognizing over twenty emotional states. So it can tell when you cry, when you laugh, and all these kind of things. It's supposed to come with long term memory as well, and roughly five hundred milliseconds of interaction response time and largely local processing for this latency of course. UBTEC reported because this has just been released already Over thirteen thousand orders at launch, but it's not independently verified. So we will see when they get delivered and how this plays out. But you can see videos of these robots. They look kind of realistic. I mean you can still tell it's not a human, of course, but they are really pushing the human like realism there. And you can actually buy them right now. I didn't find them on their non-Chinese site. You cannot buy them there, but they're like third party vendors where you can buy them. So that's UBTech. Another company which has been in this space for a while is a Real Botics and they produce highly realistic AI humanoids as well. One example is their model called ARIA and it combines silicone, face, facial tracking, also conversational AI, and roughly forty motorized degrees of articulation. So the face takes a lot of engineering work for them obviously because that's where the magic happens, right? If you talk to someone and they don't respond to you at all then There's no connection. But you can already buy it. There's an entry level that starts at around $20,000, full body one at around 125,000. So these are actually more expensive than even state-of-the-art humanoids sometimes. And what's interesting about them, because they kind of started with this romantic narrative, I think, but then the company structure changed. And robotics explicitly only focuses on commercial and service humanoids. So you know, more like professional. Human-like appearance, not really this romance or companionship type of thing. And that's why they separated the business into Intima. So Intima LIC became a separate company from them for adult wellness and companionship. And Intima, on the other hand, owns 86% stake in the real doll business. And Real Doll is also some sort of long-running company that Ships these AI enhanced silicon companions. They also have like customizable bodies. They have a proprietary skin technology, for example, but they're still pretty heavy and not really that mobile. I wouldn't call it a walking humanoid yet. Also, it's a very niche market because these are quite expensive. They said in June 26 they were actively shipping a batch of 12 units. Then we have another company which is called Ahead Form, and that is a Chinese startup. And they're developing One of the most visually human-like robotic hats and upper bodies. So they're solely focused on that, but they really do great work. They have some amazing videos on their website and they have a couple different models already. One is called the Origin F1, then the Origin M1, and then the ELF HUAN. And there's also a 2.0 version of that. And the Origin F1 has 26 facial degrees of freedom, multimodal vision and voice, face recognition. So imagine you really talk to a robot. With another person maybe and then it says like, Hey look, it's you Leo. Hey look, that's you, Oscar. You know, I can talk to you, it can differentiate, and then maybe even has memory for each person and it can relate to that. It can be a really deep connection there. They have something called the Origin M1 as well, which has 30 movable facial muscle units, micro expressions, eye tracking, and skin advertised around human body temperature. So we they really want to maintain skin that it has actually the temperature of a human. So it feels more natural. Then they have a elf, which is called Shuan, with also over 30 facial degrees of freedom. And they tease a 2.0 version on the website already. It's more like a inquiry reservation type of thing rather than just like a transparent pricing. So it's probably around 50 to 150,000 for these. And then we have Zomnia Lab. And these are quite new. And I don't really know how legit they are, but it's kind of like a new model that they are working on, the Model L. And L stands for love machine. And this is presented as a full-size intimacy robot whose personality, memory, and relationship are intended to develop over time. So this is really going for this full fantasy. Okay, I'm pretty much marrying a robot right now. They have some provisional specification, which is approximately 170 centimeters, 22 kilograms, so it's quite a light, a carbon fiber skeleton, 22 degrees of freedom. Soft compliant joints, pressure, contact sensing, localized warmth, customizable appearance. Now, what's interesting about them is they have a website and there's like a blurred layer on top of the actual website content. So I'm not sure how much of this content is just what I was able to scrape using an LM and how much is actually legit. So I just want to say that. But apparently they have a couple different models and they will take deposits for reservations from around $300 for the Model L. 500 for the Pro and 2000 for the Pro Max. So that's another one. Then we have Love Sense AI doll, and this is a companion doll. Much more affordable at around $4,000 to $8,000. Full silicon body, approximately $155 to $159. 33 kilograms, articulated posability, facial movement, touch sensing, persistent conversational memory. You know, all these features that we already talked about. They mostly go for the same things, obviously. But this one is fundamentally more a posable AI intimacy doll than like a real humanoid. We have another couple mentions. One is Jigglyjoy X Robots. It's a company working on this lifelike skin. So they really try to get skin to feel exactly like human skin. We have a couple of different Chinese doll makers that add AI. And then interestingly, I also found out about a cyber brothel ⁓ in Berlin. It's actually here in Berlin.
speaker-0: How convenience.
speaker-1: I went on their website and I looked at it and it's basically yeah, a brothel but they have these dolls, you know, they have these kind of dolls and people are apparently going there and having a lot of fun with them. So it's already taking place. I mean whoever is like still debating this, it will come and maybe there will probably be more people going for this type of situation ship or relationship. Who who knows, you know?
speaker-0: Yeah, definitely. Let let me know if you visit the Berlin establishment and you can make a vlog out of it.
speaker-1: Yeah, I'll I'll take you with me and then we we will have a podcast there with ⁓ with the founder maybe.
speaker-0: That could be actually an interesting thing. Yeah. I I'm curious about when will we have the first IPO of, you know, a robot companion startup. The day it happens, ⁓ we will definitely have massive adoption of robots in our daily lives, even beyond the companionship thing, I think.
speaker-1: I mean UbiTech is pretty big. I'm not sure if they are had already an IPO.
speaker-0: ⁓ yeah, they had a they had an IPO, UBTEC.
speaker-1: ⁓ yeah. So that's one. I mean then we got one already.
speaker-0: Yeah, actually UbiTech is already public, but their share kind of went down a little bit after Unitries IPO. I think UbiTech was not in the robot companion business at first, right? They were a traditional robotics company.
speaker-1: Yeah. I mean they just released this whole entire new section.
speaker-0: It's a new thing for them. So maybe Unitree will get into it as well. I don't know. We will see. It doesn't seem to be part of their plan, but anything is possible in this space and everything moves super fast. And it's actually a good way to talk about the Unitree IPO that you have probably heard about. It's been literally everywhere. So Unitree priced its share at twenty two point four dollars on August sixth. And they went public on August nineteenth, which is yesterday at the time of this recording, and they were aiming to raise nine hundred and four million dollars. And what happened is that when the markets opened, the company was up six hundred and thirty percent. So are massively oversubscribed. It was valuing the company at sixty-six billion dollars. Then it went down a little bit and it closed its first day up four hundred and eighty eight percent. at a fifty three billion dollars valuation. So yeah, I mean it's a ballpark, right? Because depending on how you look and the exchange rate, some sources say like it's fifty one billion, some other fifty three billion dollars, but you you get the idea. It's ⁓ it's an over fifty billion dollars valuation in a public market. If we look at multipliers, it's two hundred and ten times their sales, one thousand three hundred times their earnings, which is massive.
speaker-1: Not bad. So
speaker-0: You know, it's open AI anthropic kind of level of private VC valuation because none of this company is public yet. They were oversubscribed ⁓ more than eight thousand times, which means if you were connecting to your train desk and wanted to buy a unitary share, you were getting one in five thousand and five hundred and twenty-six chance of winning one. The stock was so oversubscribed that they went from offering 20% to retail investors to offering 30% in the end. So yeah, you had to get pretty lucky just to get one. ⁓ who profited the most from the IPO? Obviously it's Wong Xing Zing, the CEO and founder, and also Mae Chuan, which is ⁓ very large Chinese tech companies that does delivery. And they made more than 70x what they invested in Unitry, more than five billion dollars. Overall. If you have to compare the Star Market Composite, which is the Chengai market aggregate, was off six point wide percent. At the same time, Unitry was up five hundred percent. We have to talk a little bit about their products. What do they do exactly? So they shipped a bit more than five thousand humanoids in twenty twenty five. It is the company in the world that shipped the most. And we are expecting much stronger numbers in 2026, not only from Unitry, but also from other companies. They went from selling less than 10 in 2023. To around 400 in 2024 to 5,000 in 2025. So that's a considerable growth. Revenue share went from 1% to 51%. And in dollars it's approximately $130 million per year. They also sell a lot of quadruplets. They have been doing quadruplets for nine years actually. And they sold more than thirty-three thousand. It is more than forty percent of their main business, approximately fifty-seven percent of their growth margin. So Humanoids is what gets the hype. Quadrupeds are what makes their business stable and real world. And they also make components actually. Even though in 2022 components were 18% of their business, it fell down to 6% in 2025. So I'm guessing Unitree doesn't want to sell components that much anymore. They just keep doing it because they make everything in-house. Because like I said, they make every critical subcomponent themselves and they have their in-house quasi direct drive, which is an actuator made. of a motor paired with a low ratio planetary gearbox. And most of the industry otherwise is more in favour of the strain wave gearbox. Unitree builds also harmonic reducers, but most of their joints are you know crazy direct drive. The reason for that is they are faster to make and faster to iterate over from an engineering perspective. And they are back driveable, which means if you control let's say a robot arm with such an iTrader, you can yourself push it the other way and you know the motor will be fine, which is not possible with other type of actuals. The caveats of this is that they make a lot of heat and they consume a high amount of poles. Obviously the two go hand in hand, right? Heat distribution is, you know, an issue like we mentioned with Tacta. And also in the previous episode we talked about someone performing surgery with two unit free robots. And one of the issues they had, even though the attempt was successful, is that robots were overheating quite a lot. So if you look at the history of the evaluation, on November 2, 2023, the Ministry of Industry and Information Technology in China issued the guiding opinions on the innovation and development of humanoid robots. So it's basically a piece of public Chinese industrial policy. And they were directed at Developing the market and manufacturing for humanoid robots in the country. The targets were in 2025 they had to have a preliminary innovation system, breakthroughs in the brain, cerebellum, and limbs, whatever that means, secure supply of core components, batch production at scales, and two, three globally influential ecosystem enterprise and two to three industrial clusters. And I'm guessing they already have that because you mentioned Ubitech, which is Chinese. already public manufacturing humanoids unitree and when it comes to industrial cluster I don't know where UbiTech is from but Unitree is from Hengzhou in China close to Shanghai and ⁓ there is also a strong cluster in humanoid robots in Shenzhen obviously so their targets were successfully reached in this case.
speaker-1: Yeah, they ⁓ Ub Tak is in change it.
speaker-0: Mm, makes sense. So yeah, we have Hangzhou and Shenzhen for clusters. By twenty twenty-seven it was also aiming at having two modules integrated into manufacturing supply chain. And ⁓ this is already something we see in China. We've seen the edgy bot stream in a factory, and this was earlier this year in twenty twenty-six. This is definitely happening by twenty twenty-seven. And what's interesting is that before that, in twenty fifteen, they had already made robotics one of the ten priority sectors in 2016 to 2020, they made a robot industry development plan and in 2021 the five year robotics plan put further emphasis into developing the robotic industry in China. And Unitary really rode that wave perfectly when it comes to getting public funding. Because the Ministry of Industry and Information Technology in China set a ten thousand unit human deployment target by end of twenty twenty six. And since Unitary alone is shipping more than five thousand, They are probably close to target, but we don't know for sure. And in China there is a really strong central government, but also giving a lot of freedom to all the states and local region, which all came up with their own public fund. And that's what Unitary took advantage of. Unitree had a pretty modest seed and they were largely unknown outside of quadruple robot research and locomotion. Basically everything robot leg related, right? Then They had Series A after the MIIT guidance, Series B in twenty twenty four, Series C in late twenty-four and early twenty twenty-five in two parts because they had a part with public money and a part with private fund. Apparently you cannot mix the both. And now they have an IPO at fifty-three billion. So it means that in 14 months their valuation went 30x, while their revenue went three hundred and thirty-five percent up. So You can see the pattern where there is like a central state designation of the priority. The provincial funds then invest in the relevant companies. And after that, private capital follows the signal. And we definitely see that with Unitary right now, live. We've seen that with BYD, for example, in the past, but it is becoming even more clear with Unitary. For sure. What's interesting is that Agility Robotics is looking apparently at a spec at around four billions with
speaker-1: Yeah, those are crazy numbers.
speaker-0: Around 300 million committed in pre-orders and multi-site deployments. They haven't released numbers on profitability or revenue, but we can see that Unitree closed its IPO at 13 times agility's valuation, right? Another interesting thing is who actually buys one? For humanoids, it's mostly research and education at 74%. After that, the rest is around commercial and consumer industry application. And ⁓ even though Humanoid revenue is clearly tied to defined industry or job because later down the line that's where the revenue will come from. See my analysis estimate that there are 250 Unitary humanoids in active labor application against more than 5,000 shit in 2025. And Unitary has apparently a 32.4% global market share in the humanoid market. And they still have a market share of over 60% on quadruped. Pretty strong position. Well, it's already oversubscribed, so I don't think I can right now, but maybe later. I don't know. But Unitree is a profitable company with over 30 million dollars in profit last year. So this is also why I think everybody is like betting on it. They announced what they intend to do with the money they raised through the IPO. So three hundred million dollars will go to embodied large models, roughly a hundred million per year.
speaker-1: So you're gonna buy some unit stock?
speaker-0: So Reunitary is definitely investing heavily into world models and robot embedded models. They don't want just to make the hardware, they also want to make the software. And they have an interesting architecture ⁓ running you know a VLA and a world plus action model in parallel, but they haven't said too much on it at this time. It's you know news from earlier announcements. They are also investing a hundred and sixty four million in ⁓ robotics body research and development. Also investing a bit less in new products and manufacturing base. And since the IPO was oversubscribed, they made they raise more money than they were intended. So they have 254 million unallocated right now. So let's see what they do with that. Maybe robot companion department. I don't know. And also recently, right before the IPO, Unitree released the Superman demo that you have probably seen. Which is like a demo of a United robot jumping super high. like a base vertical jump and also running at approximately twelve point six meters per second, which is more than Houston Bolt, for example. Again, nothing is independently verified. It's a company marketing video right before the IPO. The robot has no hands, no payload, but the robot games are happening, so we'll get actual numbers on a hundred meters dash later on. And it is still a pretty impressive you know feat of engineering. And it shows that unitary is not just Exiting with the IPO and then being done with it, they are just doubling down on R and D and getting into what's next for humanoids. ⁓ also we mentioned that Unitree was part with NVIDIA of the Isaac growth reference humanoids. But basically Nvidia ships the software and Unitree ships the hardware to get like a baseline humanoid to do research on. It ships with a Unitree H2 and a lot of Nvidia hardware for embedded compute. and NVIDIA software, but since then there has been the import ban and they have not communicated on what's happening with this project and how can it you know go further. I'm guessing Nvidia is going to partner with US companies and Unitree investing into developing its own models. So I'm guessing the two are going to kind of split up. A little bit more about the world humanoids robot game. So they are running between August twenty two and August twenty sixth at Beijing National Speed Skating Oval Stadium. They have more than six hundred teams, more than two thousand robots and sixty six countries across six continents, and fifty-one events, and eight of the events are data rest hands competition, which I find pretty interesting. What's also really cool this year is that the hundred meter dash is fully autonomous, no teleoperation. So again I'm really curious to see what kind of time are the robots able to make on this. And more importantly, are they going to go faster than the fastest humans ever, who is use and bolt, right? It makes me wonder a little bit overall is that does Unitree need the US market? Because for example, ⁓ DGI, which is the brand of our microphone right now, is banned from the US and they keep, you know, ⁓ going around this band by rebranding in weird and funny ways. And at the same time, the US is thirteen percent of Unitree's business. So do they really need it or would for them just focusing on Europe, Asia, Africa? Would be enough for them. ⁓ I don't know what's your take on that.
speaker-1: I mean, I've I've absolutely no idea, but I saw someone in the comments writing that there's he learned from some shop who's already starting like manufacturing robots in the US. And I'm not really sure if there's gonna be like some licensing deal going on and to take more control. And I've absolutely not verified this claim so far, but I think this looks like an interesting approach that I could see working, some sort of licensing and then local manufacturing.
speaker-0: Yeah, yeah, definitely. So yeah, it was a fun episode to research and robotics just keeps getting exciting as a space. So thank you everyone for watching. Please consider subscribing to our channel and to our Twitter. And we will see you again next week in the Robotics Tag podcast. My name is Leo. Bye bye. See ya.
speaker-1: My name is Oscar.
